Study On Prediction Of Somatic Cell Count In Raw Milk And Inflammation Grade Of Dairy Cattle Based On Dielectric Spectrum

Somatic cell count(SCC)in cow's milk is an important indicator for health status of dairy cattle and milk quality,and it reflects the inflammation grade of dairy cattle.The traditional methods to detect SCC in raw cow's milk are mostly laboratory methods,which are time-consuming,expensive,and not suitable for on-line detection.Radio frequency/microwave dielectric properties method is widely used for liquid food quality analysis in recent years since it has advantages of high precision,simple operation and on-line detection.In this study,an influence law was studied on dielectric properties of raw cow's milk over the frequency range of 20~4500MHz and SCC range of 0.7~497.1 million per milliliter,and inflammation grade(negative,weak positive and positive)by using coaxial probe technology at 25?.A mathematical model was built for describing the relationship between SCC and dielectric properties of raw cow's milk.Cheometrics methods were used to detect SCC / inflammation grade based on dielectric spectra.The main works and conclusions are as follows:(1)A influence law was studied between dielectric properties of cow's milk and frequency range of 20~4500MHz at 25?.Over the investigated frequency range,the ?' of cow's milk decreased while the ? " decreased first,then increased with increasing of frequency.The ?" had a minimum value at about 1800 MHz.(2)A linear relationship was studied between dielectric properties of cow's milk and SCC over the frequency range of 20~4500MHz at 25?.Over the obtained SCC range of cow's milk,?' decreased while ? " increased with increasing of SCC.The ralationship between ?' and ?" of cow's milk and SCC could be accurately described by a linear equation with one unknown.However,at a given frequency over the whole frequency range,the linear analysis result for 153 dielectric parameters and SCC showed that the relationship between dielectric parameters and SCC was very weak.(3)Models to predict SCC in cow's milk were established based on dielectric properties over the frequency range of 20~4500MHz at 25?.All samples(153)were divided by Sample Set Partitioning Based on Joint X-Y Distances(SPXY)method.The characteristic variables were selected by Successive Projection Algorithm(SPA).Partial least square regression(PLSR)and generalized regression neural network(GRNN)and support vector regression(SVR)and extreme learing machine(ELM)and least square support vector machine(LSSVM)models were built using the full dielectric spectra and selected characteristic variables by SPA.The results showed that LSSVM model coupled with SPA had a high prediction precision in predicting SCC in raw cow's milk,with Rc,RMSEC,Rp,RMSEP and RPD of 0.850,0.294 log SCC,0.927,0.173 logSCC,and 2.375,respectively.(4)Models to predict inflammation grade of dairy cattle were made based on dielectric properties.All samples were divided samples by using Kennard-Stone(KS)method,under the full spectra and characteristic variables selected by SPA,partial least square discrimination analysis(PLS-DA)and support vector machine(SVM)and extreme learing machine(ELM)and error back propagation(BP)models were applicated to predict inflammation grade of dairy cattle,respectively.The results showed that BP model coupled with SPA had a high precision with the inflammation grade of dairy cattle,the accuracy of prediction set was 94.12%,among them,the accuracy of negative,weak positive and positive prediction set were 87.50%,95.24%,and 100% respectively.The study results suggest that combination of the coaxial probe technique and chemometrics can be used to predict SCC in raw cow's milk and inflammation grade of dairy cattle.This research provides an experimental basis for development of rapid raw cow's milk SCC and inflammation grade of dairy cattle based on the dielectric properties,which has important meaning to supervise the quality of milk.